identifying geochemical anomalies - crc geochemical anomalies.pdf · identifying geochemical...
Post on 20-Mar-2019
Embed Size (px)
IDENTIFYING GEOCHEMICAL ANOMALIES K.G. McQueen CRC LEME Department of Earth and Marine Sciences Australian National University, ACT 0200 Ken.McQueen@canberra.edu.au Geochemical anomalies are geochemical features different from what is considered normal. They can be the result of: 1. unusual or uncommon processes concentrating particular elements (e.g. an ore-forming
process, weathering and element dispersion from an unusual element concentration such as an orebody);
2. element accumulation or concentration by common processes acting over long periods (e.g. scavenging and concentration of certain elements by ironstones, ferrruginous regolith or manganese oxides);
3. artificial contamination of sites or samples; 4. analytical noise or error (e.g. poor precision of the analytical method, particularly for
element concentrations close to the detection limit). Traditionally, geochemical anomalies have been identified by setting threshold values, which mark the upper and lower limits of normal variation for a particular population of data. Values within the threshold values are referred to as background values and those above or below as anomalies. In mineral exploration interest is generally in positive anomalies, on the assumption that ore deposits and their weathering have increased element abundances above normal crustal levels. However, negative anomalies can also be important, for example where they reflect depletion in some elements during host rock alteration accompanying ore formation. Statistical methods have been widely applied to interpret geochemical data sets and define anomalies. Such methods need to be used cautiously because of the particular characteristics of geochemical data. Geochemical data sets seldom represent a single population or distribution, the data are typically spatially dependent and at each sample site a range of different processes have influenced the element abundances measured. The data are also imprecise due to unavoidable variability in sampling methods and media and the level of analytical precision. As a result no single universally applicable statistical test has been developed for identifying anomalies. Statistical investigation should use a range of techniques to explore the nature of geochemical data before selecting anomalous values (e.g. Reimann et al., 2005). Univariate statistical methods for investigating geochemical data Univariate statistical methods (i.e. involving observations with only one variable) can be used to organise and extract information from a data set of values for a single element (e.g. gold analyses for a group of samples). A first step is to examine the frequency distribution (spread of values) of the data set using frequency histograms, frequency plots or cumulative-frequency plots. This can help identify the type of distribution of the data, presence of multiple populations and outliers in the distribution. Box and whisker plots are another convenient way of examining the frequency distribution of a data set and for comparing the frequency distributions of multiple data sets. This type of plot shows the median (middle value or 50th percentile), a box with upper and lower hinges (or limits) defined by the 75th and
25th percentile values respectively (i.e. upper and lower quartile values), an inner fence defined as 1.5 timed the length of the box (the interquartile range) towards maximum and minimum values and whiskers extending to the maximum and minimum values (Tukey, 1977). The central box will contain 50% of the data. Values below the whiskers are considered outliers and values more than 3 times the interquartile range (length of the box) from the box hinges are referred to as far outliers (Figure 1).
Figure 1. Example of a box and whisker plot with corresponding frequency distribution. In the past, a simple way of statistically defining an anomaly in a single population of normally distributed data has been to consider values outside 2 standard deviations from the mean (statistical average) as anomalous (Hawkes and Webb, 1962). In other words the threshold or limit of normal variation was set at 2 standard deviations from the mean (Figure 2a) and the anomalous values taken as the top 2.5% of the population (positive anomalies) and the bottom 2.5% of the population (negative anomalies). This is somewhat arbitrary, and rarely do geochemical data fit a normal distribution pattern (they are typically positively skewed with a long tail towards higher values, Figure 2b). Commonly there is more than one population of data present in a geochemical data set (Figure 3). For example, samples collected of different media or of the same media derived from chemically different host rocks will contain multiple populations, each with their own threshold value. Further, anomalies of interest are defined by outliers that are not part of the background population.
Figure 2. Examples of frequency plots for (a) normal (at left) and (b) skewed distributions.
Figure 3. Example of frequency histogram and frequency curves for a data set with multiple populations. Another common approach has been to transform geochemical data to a normal distribution pattern and then apply normal parametric tests. It has been argued that many natural data, including trace element abundances, approximate a lognormal (log10) distribution, so a simple log transformation has typically been applied. Again this is only valid for a single population of data. Also it has been shown that most geochemical data sets lie between normal and lognormal distributions (Reimann and Filzmoser, 2000). An alternate method for handling skewed data is to set the threshold at 2 Median Absolute Deviations (MAD) from the median (middle or 50th percentile value). The median will lie away from the mean and the skewness of the data (typically to a lower value than the mean) and extreme values (outliers) will have less influence. The medium absolute deviation is defined as the median value of the absolute deviations from the median of all the data (Tukey, 1977). The MAD approach is best applied when the data contain less than 10% outliers. Background and anomalous values are commonly established empirically. During geochemical exploration programs, orientation surveys or case studies that compare typical background materials and sites with materials from areas of known mineralisation can be used to establish thresholds. This approach assumes that all the natural variability is covered in the orientation survey. It may miss very subtle anomalies or anomalies and element association patterns associated with a different or unknown style of mineralisation. A common deficiency with case studies of geochemical patterns around known mineralisation is insufficient coverage of the surrounding background. Multivariate statistical analysis of geochemical data Geochemical data sets are inherently multivariate (i.e. they generally have more than one variable for each sample, such as a large number of contained and associated elements). Geochemical anomalies are commonly expressed in more than one element. This is because the source or process that has generated the anomaly commonly has an association of elements. Different ore deposit types typically have specific element associations of target and pathfinder elements. A target element is the commodity being sought and a pathfinder element is one that accompanies this element, but may be more widely dispersed or easier to detect. Element associations can be used to advantage by taking a multi-element approach to anomaly detection. Multi-element analysis can also identify other non ore-related associations, such as those generated by normal regolith processes or the result of anthropogenic contamination.
A range of multivariate statistical methods can be used to assess the relationships within multi-element data sets. These methods commonly include: 1. scatter plots (bivariate plots comparing pairs of elements); 2. correlation matrices (using linear regression to test the correlation between pairs of
elements); 3. cluster analysis (hierarchical grouping of elements in a data set with differing degrees of
correlation of their abundance); 4. principal component factor analysis (useful for grouping elements into associations); 5. discriminant analysis (a method of optimising the distinction between two or more
populations of samples). The main difficulty in assessing multi-element data is multi-dimensional visualisation. A number of techniques can be used to help, including the use of multi-element spider diagrams, which plot values for a range of elements (connected by lines) in each sample. Typically these plots involve normalising the data to a reference sample. They have been widely used for example in comparing REE (rare earth element) data from different samples (see Rollinson, 1993). Linked scatter plots, in which particular samples can be identified across a number of bivariate element plots, are another convenient way for visually identifying samples with unusual multi-element characteristics. Triangular diagrams and computer generated rotatable 3D plots can be used to visually examine data sets for three elements at a time. Two-dimensional dendograms produced from cluster analysis are a simple way of assessing multi-element associations. A number of software programs are available for univariate and multivariate statistical analysis of geochemical data and visualisation of the results (e.g. ioGAS). Where element association are well known for particular geological materials including ore deposits, suites of these of elements can be statistically combined to detect anomalies. Simple methods that have been used include addition or multiplication of different element concentration or weighted values